Here are my responses to Suzanne's questions and comments: Questions, concerns, observations
1. Is there a difference between scale and continuous variables? Are these really synonyms? These are synonymous terms. I like “continuous variables” because I think it accurately reflects the data. It also doesn’t become confused the Likert “scale”. Most often though “scale” is used…primarily because that is what SPSS uses as a descriptor of level of measurement.
2. As I have begun to play with my data set in SPSS, I realize that it is important to double check the level of measurement (scale, ordinal, or nominal) assigned to each variable before working with the dataset. Many of the demographic variables were listed as scale, when in fact they were actually nominal. Yes, be very careful. I frankly think that the majority of people who construct databases ignore that column. I believe that the default is “scale” and so remains as such if was not changed by the person entering the data
3. I am struggling to figure out how to classify the level of measurement of the data gathered from the Likert scales in survey questions 12-25. When I look at my data set, these were all automatically assigned as scale variables. My initial inclination was to consider them ordinal data.
Yet Kinnear raises the question whether ratings can be considered scale data. Kinnear notes that some say that the use of reference or anchor points, found at the far ends of the scale, provide a form of reference for raters’ judgments, so they can be considered interval data. He also notes that others consider the individual rankings within the scale as merely ordered categories, so that 100 respondents are merely generating 100 sets of ranks with ties (multiple counts in some ranks?) , i.e., the ratings themselves are merely ordinal data (p. 2). Vogt notes that ordinal data are often treated as continuous variables when there are many ranks in the data, and as ordinal when there a just a few (p. 61).
Based on my data set, I do not think we can assume that the intervals between the 5 response categories are equal, although they do appear to have a rank order. The inclusion of the “I choose not to answer this question” option in the scale also complicates things, because this is not ranking, per se. The best advice I saw came from Jamison who suggests that decisions regarding levels of measurement should be considered at the design stage and must be addressed in the methodology section of their research (p. 1218).
Your questions demonstrate great thoughtfulness! Good for you. I agree with you that Likert scales are ordinal. I do not think that there are any sound reasons to assume that there is an equal difference between: extremely dislike, dislike, neutral, like, extremely like. In fact, there is a good reason to believe that people who pick the middle 3 options are much closer together than those who select either “extreme” option. I recommend that you treat all Likert scales as ordinal variables.
4. Preparing data for analysis. One stipulation for using this data was that I maintain the confidentiality of the participating institutions by eliminating any indentifying information that was specific to a course, instructor, program, or SUNY campus. I have removed the student comments field in the database and have started to recode the campuses into categories: community colleges, comprehensive colleges, technical colleges, specialty colleges and university centers. Good!
5. What are my options for dealing with the Likert response “I do not wish to answer this question.” This appears everywhere in the survey. If I delete these responses, each survey question will have a different denominator. Is this a problem? Or is this a matter of documenting this as a footnote? For now, include this response as an option. Don’t just delete these right away. There may be some interesting things to learn about this. I recommend coding it “99: RA (refused to answer)”. The reason that I selected that number is to keep a consistent numbering for all RAs for your ease in coding. Assume that you have the gender question: male, female, transgendered, refuse to answer. You would probably code male=1, female=2, transgendered=3, RA=4. Now, the next question may be about race: White=1, Black=2, Asian American=3, Hispanic=4, Native American=5, RA=6. Then another question has a 7 item Likert scale (so NA=8). I don’t know about you, but I like to make life easy for myself. If I set up one number and use it for every RA in the database, I don’t have to remember which it is…and sometimes you might see an interesting pattern develop in your data that you might miss without the same coding. Does this make sense?
Great work on the first learning journal! I think that writing a weekly journal may be a valuable resource for at least the first half of the course, if not for all of it.
Questions, concerns, observations
1. Is there a difference between scale and continuous variables? Are these really synonyms?
These are synonymous terms. I like “continuous variables” because I think it accurately reflects the data. It also doesn’t become confused the Likert “scale”. Most often though “scale” is used…primarily because that is what SPSS uses as a descriptor of level of measurement.
2. As I have begun to play with my data set in SPSS, I realize that it is important to double check the level of measurement (scale, ordinal, or nominal) assigned to each variable before working with the dataset. Many of the demographic variables were listed as scale, when in fact they were actually nominal.
Yes, be very careful. I frankly think that the majority of people who construct databases ignore that column. I believe that the default is “scale” and so remains as such if was not changed by the person entering the data
3. I am struggling to figure out how to classify the level of measurement of the data gathered from the Likert scales in survey questions 12-25. When I look at my data set, these were all automatically assigned as scale variables. My initial inclination was to consider them ordinal data.
Yet Kinnear raises the question whether ratings can be considered scale data. Kinnear notes that some say that the use of reference or anchor points, found at the far ends of the scale, provide a form of reference for raters’ judgments, so they can be considered interval data. He also notes that others consider the individual rankings within the scale as merely ordered categories, so that 100 respondents are merely generating 100 sets of ranks with ties (multiple counts in some ranks?) , i.e., the ratings themselves are merely ordinal data (p. 2). Vogt notes that ordinal data are often treated as continuous variables when there are many ranks in the data, and as ordinal when there a just a few (p. 61).
Based on my data set, I do not think we can assume that the intervals between the 5 response categories are equal, although they do appear to have a rank order. The inclusion of the “I choose not to answer this question” option in the scale also complicates things, because this is not ranking, per se. The best advice I saw came from Jamison who suggests that decisions regarding levels of measurement should be considered at the design stage and must be addressed in the methodology section of their research (p. 1218).
Your questions demonstrate great thoughtfulness! Good for you. I agree with you that Likert scales are ordinal. I do not think that there are any sound reasons to assume that there is an equal difference between: extremely dislike, dislike, neutral, like, extremely like. In fact, there is a good reason to believe that people who pick the middle 3 options are much closer together than those who select either “extreme” option. I recommend that you treat all Likert scales as ordinal variables.
4. Preparing data for analysis. One stipulation for using this data was that I maintain the confidentiality of the participating institutions by eliminating any indentifying information that was specific to a course, instructor, program, or SUNY campus. I have removed the student comments field in the database and have started to recode the campuses into categories: community colleges, comprehensive colleges, technical colleges, specialty colleges and university centers.
Good!
5. What are my options for dealing with the Likert response “I do not wish to answer this question.” This appears everywhere in the survey. If I delete these responses, each survey question will have a different denominator. Is this a problem? Or is this a matter of documenting this as a footnote?
For now, include this response as an option. Don’t just delete these right away. There may be some interesting things to learn about this. I recommend coding it “99: RA (refused to answer)”. The reason that I selected that number is to keep a consistent numbering for all RAs for your ease in coding. Assume that you have the gender question: male, female, transgendered, refuse to answer. You would probably code male=1, female=2, transgendered=3, RA=4. Now, the next question may be about race: White=1, Black=2, Asian American=3, Hispanic=4, Native American=5, RA=6. Then another question has a 7 item Likert scale (so NA=8). I don’t know about you, but I like to make life easy for myself. If I set up one number and use it for every RA in the database, I don’t have to remember which it is…and sometimes you might see an interesting pattern develop in your data that you might miss without the same coding. Does this make sense?
Great work on the first learning journal! I think that writing a weekly journal may be a valuable resource for at least the first half of the course, if not for all of it.
Well done!
d